A Systematic Literature Review on Word Embeddings

This article presents a systematic literature review on word embeddings within the field of natural language processing and text processing. A search and classification of 140 articles on proposals of word embeddings or their application was carried out from three different sources. Word embeddings have been widely adopted with satisfactory results in natural language processing tasks in general and other domains with good results. In this paper, we report the hegemony of word embeddings based on neural models over those generated by matrix factorization (i.e., variants of word2vec). Finally, despite the good performance of word embeddings, some drawbacks and their respective solution proposals are identified, such as the lack of interpretability of the real values that make up the embedded vectors.